Memory-Augmented Multimodal Video Retrieval With Efficient Long-Term Visual Context Modeling. JDIAS 2026, 1 (1).